Automated Long Answer Grading (ALAG) presents unique challenges compared to shorter forms like ASAG or essays due to their complex, fact-based structure. This research introduces the RiceChem dataset, derived from a college chemistry course, to specifically address these challenges.

Summary:

• Rubric-based Assessment: Models align student responses with a detailed rubric for accuracy.
• Transfer Learning Utility: Leveraging MNLI model insights improves grading accuracy significantly.
• Performance Benchmarks: Tests against large language models like GPT show the unique difficulties in grading long answers.
• Future Research: Encourages further exploration into automated systems for educational settings, potentially transforming how students are assessed.

This groundbreaking study not only pioneers rubric-based ALAG but also sets a new benchmark for educational AI applications.